Lire en français

Some organisations are already cutting headcount in anticipation of agentic AI’s gains. In many cases, processes aren’t mature and agents aren’t reliable enough at scale.

That sequencing feels risky. Cutting headcount before the technology and the organisation are ready means paying upfront for what AI hasn’t delivered yet: expertise that’s hard to rebuild — especially critical in regulated environments — and errors on the most complex processes.

Many leaders also believe a well-crafted “AI transformation” communication plan will be enough to carry the change. It rarely is.

In 2023, when generative AI reached regulatory affairs, my first question wasn’t technological. It was managerial: how do I keep my people employable?

Here’s how I went about it.
I put the subject in front of my teams early, rather than waiting for it to impose itself. Through workshops and brainstorming, I asked what they wanted to stop doing. Ambassadors — chosen for their clarity, their ability to bring people together, and their grasp of regulatory processes — facilitated the sessions. Real momentum built. Then we structured it: ROI per use case, selection criteria, prioritisation.

And that’s where I learned something unexpected.
The ideas that excited the room weren’t the ones that survived scoring. The winning use cases were the boring ones: repetitive, time-consuming work demanding rigour and precision but little advanced technicality. In regulatory affairs, there’s a lot of it.
At the time, disappointing. I now think it was the real output of the exercise: scoring protected us from hype — including our own.

One thing I keep in mind: a scoring isn’t final. As more tools reach the market, development effort falls — a use case ruled out today can become viable tomorrow.

Ambassadors don’t just identify use cases. They design and validate the first agents, then carry deployment within their own scope. That’s the difference between a pilot and an actual practice.

What this approach produces, fundamentally, is capacity. The framework for introducing AI is decided at company level. What I choose, within my scope, is how it’s used — and I chose augmentation. In regulatory affairs, the list of things we’d like to do and never do is long: data cleanup, systematic intelligence monitoring, harmonising practice. It’s the first thing deprioritised when capacity is short — in a mid-sized organisation like mine, but I see it in far larger ones too.
Agents don’t replace — they unlock the work that never gets done.

Competitive advantage won’t come from those who cut the most jobs betting on hypothetical gains, but from those who worked out what to automate — and that can’t be worked out without the people who do the work.

If you’ve scored your AI use cases: were the winners the exciting ones, or the boring ones?